Re: pg_xlogdump

Dimitri Fontaine <dimitri(at)2ndQuadrant(dot)fr> writes:
> Peter Eisentraut <peter_e(at)gmx(dot)net> writes:
>> I for one wonder why we even have PGXS support in contrib at all. It's
>> not documented or tested anywhere, so it might as well not exist.

> I think I did about the same comment back when cooking the extension
> patch, and the answer then was all about providing PGXS usage examples.
> Now if none of the buildfarm animals are actually building our contribs
> out of tree, maybe we should just remove those examples.

> The cost of keeping them is that they double-up the Makefile content and
> lots of users do think they need their extension's Makefile to be
> structured the same. The common effect before the extension availability
> was for people to provide extensions that would only build in tree.

> I don't want to kill cleaning up those Makefiles, but I still want to
> make a strong correlation in between that point and providing core
> maintained extensions. I don't think extensions should have support for
> being built in-tree at all.

> My proposal: paint them extension rather than contrib modules, then
> cleanup Makefiles so as to stop building them in-tree.

[ Sigh... ] Why this eagerness to fix what isn't broken?

Leave the Makefiles alone. They're not broken and they provide useful
examples, plus a sense of continuity between in-tree and not-in-tree
extensions. Any change here will likely break build scenarios that
work today --- in particular, this proposal will break building contrib
before the main tree has been installed.

If somebody wants to set up a buildfarm member that occasionally tests
PGXS building of contrib/, that's fine with me. But it isn't, and never
will be, the main build scenario for contrib/ IMO.

regards, tom lane